I stopped writing meeting minutes by hand about three years ago, and I have not missed a single agreed action since. That is not because my memory improved. It is because I let software do the remembering, and I built a routine around checking its work.

If you want the short version before we go deep: an AI meeting notes tool records or listens to your call, transcribes it, and turns the transcript into a summary with action items, usually within a few minutes of the meeting ending. For most UK professionals, Fathom is the best free starting point, Fireflies is the best value paid option if your notes need to land in a CRM, and if you work in a regulated sector such as financial advice you should look at a UK specialist like Aveni or TakeNote rather than a generic American tool. You also need to tell people they are being recorded, because in Britain the Information Commissioner's Office treats AI transcription as data processing under UK GDPR, and that comes with real obligations. Everything below is the detail behind that answer, including current prices, the tools I would avoid, and the compliance steps that most articles on this subject skip entirely.

Why Minutes Go Missing in the First Place

British working life is drowning in meetings. A survey of 1,000 UK professionals by Raconteur and Attest found that nearly one in five spends more than 10 hours a week in virtual or in person meetings, and almost one in ten spends more than 15 hours. That is two full working days for some people, before a single email gets answered.

The cost of all that talking is not trivial either. Research by Epson and the Centre for Economics and Business Research put the bill for wasted meeting time at roughly £26 billion a year for UK businesses, with office workers estimating that over half their meeting time added no value. More recent YouGov work reported by BM Magazine suggests UK firms lose around £13 billion a year to wasted managerial time, a chunk of which is meetings that produce no usable record.

Here is the bit that matters for this article. The waste is not just the meeting itself. It is what happens afterwards. Decisions get made, actions get assigned, and then everyone walks out and forgets. Widely cited memory research suggests people forget around half of what they heard within a day. If nobody wrote proper minutes, that decision effectively never happened, and you get the same conversation again a fortnight later.

Manual minute taking was always the fix, and it was always a bad one. Whoever takes the notes cannot fully participate. The notes reflect one person's attention span and typing speed. And in client facing work, scribbling while someone explains their situation is both rude and risky, because the thing you miss is usually the thing that mattered.

What an AI Meeting Notes Tool Actually Does

An AI meeting notes tool is software that captures the audio of a conversation, converts it to text using speech recognition, then uses a large language model, which is the same kind of AI behind ChatGPT and Claude, to produce a structured summary. A decent one gives you four things: a full transcript with speakers labelled, a summary of what was discussed, a list of decisions, and a list of action items with owners.

There are two fundamentally different designs, and choosing between them is the first real decision you will make.

Bot based tools join your Zoom, Google Meet or Microsoft Teams call as a visible participant. Otter, Fireflies, Fathom and tl;dv all work this way. Everyone sees "Fred the notetaker has joined" or similar. That visibility is annoying to some people, but it is genuinely useful for transparency, because nobody can claim they did not know the call was being captured.

Botless tools run quietly on your own device and capture the audio from your microphone and speakers. Granola is the best known example, and Microsoft Copilot has a mode that works without saving a recording. These feel more natural, and clients never see a robot in the participant list. The trade off is that transparency is now entirely your responsibility. Nothing on screen tells the other person anything is happening, so you have to say it out loud. More on why that is a legal requirement, not just good manners, further down.

One more thing worth knowing before you compare products. Transcription accuracy has largely stopped being the differentiator. In hands on testing published by Atlas Workspace, the top tools all achieved strong accuracy on real meeting audio, with Otter recording the lowest word error rate at 6.3 percent across eight recordings. The same testing found that marketing claims of 99 percent accuracy are usually based on clean studio audio, and that messy real world calls with overlapping speakers can knock accuracy down by 15 to 20 percentage points. So do not choose a tool because its website claims the best accuracy. Choose it based on workflow, privacy, and where the notes end up.

The Tools I Rate, With Real Prices

A note on money before the list. Almost every mainstream tool in this category is American and bills in US dollars, so the pound figures below are approximate conversions and will drift a little with the exchange rate. Where a supplier publishes UK pricing I have said so. All of them are available and fully functional in the UK.

Fathom

Best for: individuals who want the most generous free plan on the market.

Fathom gives you unlimited recordings, transcription and storage on its free tier, which no direct competitor matches. The catch, according to the detailed pricing breakdown at CheckThat, is that free users get only five advanced AI summaries a month, so heavy users end up on Premium at 16 dollars per user per month billed annually, which is roughly £13, or Team at 18 dollars, roughly £14, which unusually includes single sign on and custom vocabulary at that price. It works with Zoom, Google Meet and Microsoft Teams, and the summaries arrive fast. If you have never used one of these tools, start here. It costs you nothing to find out whether the habit sticks.

Fireflies

Best for: teams whose notes need to flow into a CRM or task system automatically.

Fireflies' free plan looks generous until you read the small print. The 800 minute storage limit is cumulative rather than monthly, so a few weeks of regular meetings fills it and older meetings start getting compressed or removed. Treat the free tier as a trial. The Pro plan at 10 dollars per user per month, roughly £8, is where it earns its keep, pushing notes into HubSpot, Salesforce and thousands of other tools, with Business at 19 dollars, roughly £15, adding sync rules and analytics. As the pricing analysis from Layer3 Labs puts it, Fireflies wins when notes must flow somewhere, not when they just need to exist. That is exactly right, and it is the reason my own team pays for it.

Granola

Best for: people who take some notes themselves and want AI to fill in the rest, without a bot on the call.

Granola runs on Mac and Windows, captures audio locally, and blends your own rough jottings with the transcript to produce enhanced notes. It is lovely to use and brilliant for in person conversations, which bot based tools cannot capture at all. Two honest warnings though. First, the free Basic plan only keeps 30 days of meeting history, a limit the pricing page does not spell out clearly, with paid plans starting at 14 dollars per user per month, roughly £11. Second, Granola uses your data to train its AI by default, and you have to switch that off yourself in the settings. For UK client work, go into settings and opt out before your first call, and remember that its transcripts sit in US hosted infrastructure, which matters for your privacy documentation.

Otter

Best for: live transcription while a meeting is still happening, and accented English.

Otter is the household name, and its live transcript view remains the best in the business. But I think it is the most overrated tool in this category right now, and I say that as someone who paid for it for two years. The free plan gives you 300 minutes a month with a hard 30 minute cap per conversation, so a single 45 minute client call loses its final 15 minutes. Worse, Otter quietly cut its Pro plan allowance from 6,000 minutes a month to 1,200 without reducing the price, which currently sits at 8.33 dollars per user per month billed annually, roughly £6.60, jumping to 16.99 dollars if you pay monthly. Business is 19.99 dollars annually, roughly £16, or a steep 30 dollars month to month. If you record a lot, Fireflies gives you far more minutes per pound. Otter still deserves a place on the shortlist for its accuracy and live view, but check the caps against your actual diary before subscribing.

tl;dv

Best for: multilingual teams and anyone who wants video clips rather than just text.

tl;dv records, transcribes and summarises in more than 30 languages, and its timestamped clips are genuinely useful for sharing a two minute moment from an hour long call. Paid plans start around 18 to 20 dollars per user per month, roughly £14 to £16. The free tier is decent for occasional use, though it is built around online video calls, so it is the wrong choice if much of your work happens face to face.

Microsoft Copilot and Teams

Best for: organisations already living inside Microsoft 365 who want notes governed by their own IT department.

If your firm runs on Teams, you may not need a third party tool at all. Basic Teams transcription is free but raw. Teams Premium at around 10 dollars per user per month, roughly £8, adds AI recaps. The full Microsoft 365 Copilot licence runs 18 to 30 dollars per user per month depending on your plan, roughly £14 to £24, and produces summaries, action items and a searchable meeting history that stays inside your Microsoft tenant. That last point is the killer feature for cautious UK organisations, because your data never leaves an environment your administrators already control. The downside is cost stacking on top of what you already pay Microsoft, and the fact that Copilot only works in Teams, so external calls on Zoom or Meet are invisible to it.

The Tools I Would Skip

Two quick warnings. First, avoid any tool that will not tell you plainly where your data is stored and whether it trains AI models on your content. If the privacy page is vague, the answer is usually one you would not like. Second, be sceptical of newer apps competing purely on price with big claims about accuracy. As the Atlas Workspace testing showed, one well known transcription service marketing 99 percent accuracy measured at 89.6 percent under independent conditions. Cheap plus overclaiming is a bad combination when the output is a record of what your client said.

Here is where UK readers need to stop copying advice from American blogs. In the United States, recording rules revolve around state consent laws. In Britain, the framework is UK GDPR, enforced by the Information Commissioner's Office, and it treats a meeting transcript as personal data because it contains names, opinions and often much more sensitive material.

That means four practical obligations, and none of them are optional.

You need a lawful basis. The two realistic candidates are legitimate interests, which is what most UK businesses rely on for routine internal recording, and consent, which is the cleaner option for client calls. The ICO's innovation advice service has addressed AI call transcription directly, noting that for consent to be valid it must be freely given, specific and informed, with a clear affirmative action from the individual. A mumbled "the robot is listening" as you dive into agenda item one does not meet that bar. If you rely on legitimate interests instead, you need a documented legitimate interests assessment, which is a short written balancing exercise showing your business need does not override the other person's rights.

You must tell people. Transparency is a core UK GDPR requirement, and a practical guide to UK GDPR and meeting transcription makes the point that covert recording in a professional setting can breach privacy rights beyond data protection law, with the ICO's Employment Practices Code saying covert monitoring of workers is justifiable only in exceptional circumstances. The fix is easy. Say at the start of every call that it is being recorded and transcribed by AI, put the same statement in your meeting invitations, and update your privacy notice to mention automated transcription. The ICO's refreshed guidance published in January 2026 also made clear that using AI to analyse recordings is a separate processing activity that must be disclosed in your privacy notice, so if you switched on an AI notetaker without updating your documentation, do it this week.

You need a contract with the tool provider. Under UK GDPR your business is the data controller and the AI company is your processor. That relationship must be governed by a data processing agreement. Every serious provider offers one, usually buried in their terms. Actually read it, and check where the data is stored. Most of these tools process in the United States, which is workable under the UK's data bridge arrangements, but you need to know and record it.

You cannot keep everything forever. Data minimisation means holding recordings only as long as the purpose requires. A sensible pattern is deleting the raw audio once the notes are checked and approved, and setting an automatic retention period on transcripts. Many tools let you configure auto deletion. Use it.

One more thing that trips people up. AI transcripts contain mistakes. They mishear names, merge speakers and butcher technical terms. UK GDPR requires personal data to be accurate, so someone should skim every transcript before it gets filed or shared. Two minutes of review is usually enough, but you need it to be a habit, not an accident.

Client Calls in Regulated Sectors

If you are a financial adviser, solicitor, accountant or anyone else whose regulator expects a documented record of client conversations, AI meeting notes are arguably even more valuable, and the tool choice changes.

Take financial advice. The FCA's Consumer Duty regime expects firms to evidence good client outcomes at the level of individual files, and advisers report spending enormous amounts of time on documentation. Research from AdvisoryAI found that paperwork and administration directly eat into client facing time for 43.3 percent of UK advisers. This is why a small ecosystem of UK specific tools has grown up. Aveni's guide to AI meeting tools for UK advisers makes the fair point that generic transcription captures the words but does not produce the structured, audit ready records the FCA expects, which is what purpose built platforms like Aveni Assist, AdvisoryAI and TakeNote are designed to do, complete with suitability note formats and audit trails. One adviser network of 70 advisers using Aveni reported saving 1,400 hours a month between them.

Even outside financial services, the principle carries. If your profession has record keeping rules, pick a tool that produces the shape of record your regulator wants, and confirm the vendor will sign a proper data processing agreement and keep your data out of its training sets. A generic consumer app is the wrong tool for a safeguarding conversation or a legally privileged call, full stop.

Four Questions That Pick the Tool for You

If the list above still leaves you torn, answer these in order and the shortlist collapses quickly.

Where do your meetings actually happen? If everything runs through Teams inside a managed Microsoft environment, look at Copilot or Teams Premium first, because your IT department already governs the data. If you bounce between Zoom, Meet and Teams, a cross platform bot tool like Fathom or Fireflies makes more sense. If a good share of your conversations happen in person, over coffee or across a desk, only a device based tool like Granola will capture them at all.

Where do the notes need to end up? If the answer is a CRM or task system, pay for Fireflies and let the integrations do the filing. If the answer is simply a searchable archive you occasionally consult, a free tier will carry you a long way.

How sensitive is the content? Routine project catchups can live happily in a mainstream American tool with the training opt out switched on. Client conversations in regulated work belong in a UK specialist with a proper audit trail, or at minimum a tool whose data processing agreement you have actually read.

How many minutes do you really record? Count a typical month before you buy. Caps, not features, are where these subscriptions quietly punish heavy users, as anyone who hit Otter's reduced Pro allowance can tell you.

Tools are only half of this. The other half is the routine, and mine has settled into five steps that take under ten minutes per meeting.

Before the meeting, the calendar invitation carries one standing line telling attendees the call will be recorded and transcribed by AI, with a contact if they object. For external client calls, I ask for a verbal yes at the start and the tool captures that consent in the transcript itself, which is a tidy little audit trail.

During the meeting, I take no notes at all beyond the occasional keyword. That is the entire point. I look at the person, I listen, and I ask better questions because my hands are free.

Within an hour of the meeting ending, I read the AI summary against my memory while the conversation is fresh. This is where you catch the transcription deciding your client's surname is a type of cheese. I correct names, fix any garbled figures, and delete anything irrelevant or sensitive that does not belong in the record.

Then the actions get moved out. A summary sitting in a notetaking app is only marginally better than a memory. Every action item goes into the actual task system, ours flow into the CRM automatically via Fireflies, with an owner and a date. This is the step most people skip, and it is the step that makes the difference between having notes and never forgetting a client call.

Finally, the raw audio gets deleted on a schedule. Approved notes and corrected transcripts stay for as long as our retention policy says, and no longer.

An Honest Reality Check

I am an enthusiast, but let me tell you what these tools will not do, because the marketing will not.

They will not understand what mattered. An AI summary weights what was said, not what was meant. The throwaway comment where a client hinted they were unhappy can end up as half a sentence in a summary, when a human note taker would have underlined it three times. Read the summaries with your own judgement switched on.

They occasionally invent things. Language models sometimes produce confident text that was never said, especially around numbers and names. It is rare with the good tools, but rare is not never, and if the note is going into a client file, it needs human eyes first.

They change the temperature of a conversation. Some people speak less freely when they know a machine is listening, and a few will refuse outright. Respect that. Have a fallback of old fashioned handwritten notes and never bury the recording disclosure hoping nobody notices, because that road ends with a complaint to the ICO.

And they will not fix a bad meeting. If the conversation was aimless, the AI will faithfully produce an aimless summary. Agendas still matter. The Raconteur research quoted a recruitment director calling Britain's meeting load a significant loss of productivity that UK employers cannot afford, and no notetaker changes that. What it changes is whether the meetings you do hold produce something durable.

What to Do This Week

If you have read this far, here is the practical path. Install Fathom or Granola today, both free, and use one on your next three internal meetings. Add the recording disclosure line to your calendar template and your privacy notice at the same time, and jot a short legitimate interests assessment, one page is fine. After three meetings, check the notes against your memory and ask one question: did anything get captured that would previously have been lost? In my experience the answer arrives fast, usually in the form of an action item you had already forgotten you agreed to.

If the habit sticks and your notes need to reach a CRM, budget roughly £8 to £16 per user per month for Fireflies or a Fathom paid tier. If you are in a regulated profession, book demos with the UK specialists instead and ask hard questions about data location, training opt outs and audit trails. Either way, AI meeting notes are one of the very few workplace AI applications that reliably pay for themselves in the first month, provided you do the unglamorous parts: tell people, check the output, move the actions, delete the audio. Do those four things and you genuinely will never write minutes or forget a client call again.